How to use Claude to build an AI visibility audit tool
Mention First's founder writes specifications, has AI agents write code, and reviews and tests the result. The product reportedly completed 26 live audits and blocks reports when deterministic answer checks fail.
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Implement [AI visibility audit specification] for [businesses and answer providers]. Use [defined deterministic checks] to validate every raw answer and block reports whenever a check fails. Present the code and testing instructions for my review, and identify any requirements that need clarification.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source poster / publisher
- VarisLatypov
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The founder reports 26 live audits, 1,818 AI answers and an average cost of $1.64 per audit. Any failed deterministic check blocks the report.
- Human involvement
- The founder explicitly says they write the specification and review and test the agent-written code.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Write the specification for the software the agents will implement.
- Have AI coding agents write the product code.
- Review and test the code produced by the agents.
The founder of Mention First describes building a product that measures how a business appears in answers from ChatGPT, Gemini, Perplexity, Claude and Grok. Their development process assigns different responsibilities to the human and the coding agents: the founder writes the specification, agents write the code, and the founder reviews and tests it. Claude Code is identified as a daily tool, used with multi-agent workflows, custom skills, hooks and MCP servers.
The product also has a stated reporting safeguard. Twenty deterministic checks run on every raw answer, and any failed check blocks the report. The source does not list those checks or explain their implementation, so the reported safeguard should not be interpreted as proof of any particular accuracy standard.
At the time of the post, the founder reported 26 live audits covering 1,818 AI answers, with an average cost of $1.64 per audit. These are reported operating results for Mention First, rather than a measured comparison of coding methods. Who it's for: founders and technical product managers who want to direct agent-written software while retaining responsibility for specifications, review and testing.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · VarisLatypov
Frequently asked questions
What does Mention First measure?
The founder says it measures how a business shows up in answers from ChatGPT, Gemini, Perplexity, Claude and Grok.
How does the founder oversee agent-written code?
The founder writes the specification, lets AI agents write the code, and then reviews and tests it.
What prevents a failed answer check from reaching a report?
The source states that 20 deterministic checks run on every raw answer and that any FAIL blocks the report. It does not enumerate the checks.
More details
Remaining steps
- Run 20 deterministic checks on each raw answer and block reports with a failed check.
About this account
AI agents write the code from the founder's specifications; the founder retains review and testing responsibilities.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=50369943. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: VarisLatypov on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · Custom skills · Hooks · MCP servers
AI visibility · Agent coding · Product development
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.